BYO Agentic Framework

Work in Progress - This repository is actively evolving with new patterns and examples.

A collection of notebooks demonstrating how to build AI agents using Llama Stack with different frameworks: pure Llama Stack Responses API, LangChain, and CrewAI.

This project explores three approaches to building AI agents with tool calling and RAG (Retrieval-Augmented Generation) capabilities, all leveraging MCP (Model Context Protocol) for tool integration.

What you'll learn:

• Build agents using Llama Stack's native Responses API (no framework dependencies)
• Integrate LangChain 1.0 agents with MCP tools
• Create multi-agent RAG systems with CrewAI
• Deploy containerized MCP servers on OpenShift

To get started, jump to installation.

Table of contents

• Architecture
• Requirements
• Installation
• Notebooks Overview

Architecture diagrams

┌─────────────────────────────────────────────┐
│         Agentic Notebooks                   │
│  (3 approaches: Primitives, LangChain,      │
│   CrewAI)                                   │
└────────┬────────────────────────────────────┘
         │
         │ API Calls
         ▼
┌─────────────────┐   ┌──────────────────────┐
│  Llama Stack    │   │  MCP Tools           │
│  - vLLM Engine  │◄──┤  - Weather Service   │
│  - Vector Store │   │  - Kubernetes API    │
│  - Responses API│   │  - Yahoo Finance     │
└─────────────────┘   └──────────────────────┘

Key Components:

• Llama Stack: Inference engine with OpenAI-compatible API
• MCP Servers: Containerized tool servers (Weather, K8s, Finance)
• Vector Stores: Document storage and retrieval for RAG
• Frameworks: Optional layers (LangChain, CrewAI) for orchestration

References

• Llama Stack Documentation
• Model Context Protocol (MCP) Specification
• LangChain MCP Adapters
• CrewAI Documentation